webis/tite-2-late-msmarco

TITE: Token-Independent Text Encoder

This model is presented in the paper TITE: Token-Independent Text Encoder for Information Retrieval. It's an efficient bi-encoder model for creating embeddings for queries and documents.

We provide the following pre-trained models encoder models:

• webis/tite-2-late
• webis/tite-2-late-upscale

We provide the following fine-tuned bi-encoder models for text ranking:

Model	TREC DL 19	TREC DL 20	BEIR (geometric mean)
webis/tite-2-late-msmarco	0.69	0.71	0.40
webis/tite-2-late-upscale-msmarco	0.68	0.71	0.41

Usage

See the repository for more information on how to use the or reproduce the model.

Citation

If you use this code or the models in your research, please cite our paper:

@InProceedings{schlatt:2025,
  author =                      {Ferdinand Schlatt and Tim Hagen and Martin Potthast and Matthias Hagen},
  booktitle =                   {48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2025)},
  doi =                         {10.1145/3726302.3730094},
  month =                       jul,
  pages =                       {2493--2503},
  publisher =                   {ACM},
  site =                        {Padua, Italy},
  title =                       {{TITE: Token-Independent Text Encoder for Information Retrieval}},
  year =                        2025
}